BYOKey PixVerse Text to Video

PixVerse is the value pick in the video API lineup: fast, comparatively cheap, and good at the flashy trending styles people actually make shorts with. If your use case is "generate a bunch of clips and pick the good ones," it beats burning Veo-class credits on every roll of the dice. This node is the bring-your-own-key way to reach it from ComfyUI, dropping straight into the same VIDEO socket the official API nodes use.

How it works. One quirk up front: unlike most vendors here, PixVerse takes your key in the API-KEY header, not as a Bearer token - so a bare key works. The node POSTs your job to app-api.pixverse.ai/openapi/v2 (the default base_url, overridable for a relay), polls until the video is ready, downloads it, and wraps it in ComfyUI's native VIDEO type. Same submit-then-poll dance as every video node in this pack, with the usual poll_interval and max_wait knobs if you need to give a job more room.

The inputs that matter:

• prompt - your text-to-video description. PixVerse has its own negative-prompt support, and it's worth using: describing what not to do (extra limbs, morphing, watermark artifacts) helps more here than on some other vendors.
• aspect_ratio - 16:9 default, with 4:3, 1:1, 3:4, 9:16 available. Pick your format before you queue; you can't recrop the API output cheaply.
• quality - 360p through 1080p. This is your cost lever as much as a quality lever. 540p is a sensible default; jump to 720p for anything you'll actually post.
• duration_seconds - just 5 or 8. No fine control here.
• motion_mode - normal or fast. Fast mode trades a bit of polish for shorter job time; useful for draft rounds.
• model - free text, defaults to v3.5. PixVerse bumps versions fairly often, so this stays flexible rather than an enum.
• template_id - PixVerse's effect templates. Leave at 0 (none) unless you have a specific template id to use; you won't know what these are from the node.
• seed - real seed semantics here (PixVerse honors it for reproducibility), so you can lock a look down once you've got a good roll.

Output: one output (VIDEO), wired to your usual video preview or save node.

Install. Same shared pack, plain clone:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git
# restart ComfyUI; find it under api/byokey/video

No models to download, no extra dependencies - it runs on ComfyUI's bundled torch, aiohttp, PIL, numpy and av.

Gotchas. Jobs still take a minute or two, so keep max_wait generous. The API-KEY header quirk means some key formats that work elsewhere (like Bearer ...) will fail here - paste the raw key. And as with any fresh API-wrapper pack, give the source a skim before feeding it a key; it's new enough that there's no community consensus behind it yet.